What's currently the 'smartest' language model?

SurpriZe@lemm.ee to Asklemmy@lemmy.ml – 40 points –

Looking for an "AI" that would tackle my day-to-day issues that are not related to programming, for example acting as a personal assistant / life coach, creating lesson plans for classes I teach at school, explaining how things work and teaching me new skills effectively, etc.

I need it to be able to consider web search options for more comprehensive answers.

Doesn't have to be free, as I'd be happy to pay if it's truly worth it.

So far I've tried:

  1. Most common options at Poe, including Claude Sonnet 3.5, GPT4o and others. The issue here is that I'm not seeing which one is actually smarter and which one hallucinates more.
  2. Perplexity
  3. Phind
  4. Gemini
  5. Bing AI

I have never had a GPT4 subscription so I might consider that if it's objectively the best option.

What can you recommend? ๐Ÿ™‚

23

You are viewing a single comment

GPT-4 is apparently the model to beat. I haven't seen all that much difference in practice between GPT-4 and 4o. I've heard various claims about various other models outperforming it (notably including Claude) but I haven't seen the claims materialize over the long haul as yet.

I have however heard that Mistral can get quite close to GPT-4, run for free locally with the right hardware, if you build up a hand curated set of around 100 query/response pairs from GPT-4 that are what you want it to do, and then fine-tune Mistral against that training set. I haven't tried it but that's what I've heard.

I'm a total layman when it comes to setting up a language model locally. Any step by step guide on how to do it? And I mostly use AIs on my Android phone, not PC. Is it possible to synchronize it between two devices?

GPT4all can do it pretty easily on a desktop with a good GPU. I think it's unlikely that anything can run locally on your phone (LLMs are notably hogs in terms of even pretty capable desktop PC resources; there's just not a cheap way to do them). You could use colab or something via your phone, and there is probably a little howto guide somewhere that shows how to do a Mistral setup on colab. It'll take some technical skill though.

You might just bite the bullet and do $20/mo for the GPT-4 subscription also. It can also do web searches, I think, although in practice it's pretty clunky the times it's tried to do things like that for me. I'm not aware of one that does the "search the web for answers and get back to me" thing really all that perfectly or smoothly I'm sad to say.

Why do the $20 subscription when the API pricing is much cheaper, especially if you are trying different models out. I'm currently playing about with Gemini and that's free (albeit rate limited).

100% right; unless you are using it a ton the API pricing is likely to be cheaper

And also, any recommendations on a specific GPT4 addon or is the base model pretty much perfect as is?

GPT-4 generally doesnโ€™t need fine tuning or anything no